Search papers, labs, and topics across Lattice.
1
0
3
Identical prompts can trigger surprisingly inconsistent and unsafe responses from LLMs under repeated inference, even in models that appear safe based on standard benchmarks.
